Other vague ideas in back of my mind make me think maybe the AIs also likely cannot do things like go log in somewhere to do something, right? Some kind of prevent them being used to attack other servers etc protection / ethics?
Yes, AI has moved from just talking to doing. The term is "AI Agent", and it can do things like read a spreadsheet, call numbers, address people by name, log into a website, change a password, etc. I've bever been to the dark web (don't use TOR) but I'm sure from what I read that you can buy such agents there. Probably still rudimentary but they are improving
daily. Automated tools for attacking servers and such have been out for years; it's the
person with the password they are attacking now.
I only recently started trying DeepSeek web interface to get some ideas how to solve the problems I am facing in a rather complex code I have been trying to write for a trading bot. My conclusion so far is that It can be useful to give you some ideas how to solve your problem but I don't think it can solve it for you. It is just like an assistant or a tool you can take advantage of.
I left DS for ChatGPT a few weeks ago. Since I live in hockey country, the speed is better and it has a useful UI that lets me tell the AI how I want it to answer. It is fairly good at understanding what I want, and it will put everything in completed files in the proper structure so I just need to drag to my extension directory. Saves a lot of time in that aspect. However every time I think I have it chugging along nicely it silently makes something "better" without telling me - I've mitigated this with instructions to ask, not assume but it still happens.
I don't think I will ever finish my SP project using only AI. I can spend a day working with it only to have to restore from backup and start again. I'm thinking I should find a willing partner who wants to share in the ranking excitement!